An Explainable AI-Enabled Framework for Interpreting Pulmonary Diseases from Chest Radiographs

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer Science

Background:

• Pulmonary diseases significantly impact the human respiratory system.
• Accurate and timely diagnosis of chest diseases is crucial for effective treatment.
• Explainable Artificial Intelligence (XAI) is vital for understanding AI-driven healthcare decisions.

Purpose of the Study:

• To develop and evaluate a CNN-based transfer learning approach for automatic explanation of pulmonary diseases from chest radiographs.
• To enhance the diagnostic capabilities for conditions including edema, tuberculosis, nodules, and pneumonia.
• To utilize XAI techniques to provide interpretable insights into disease classification, particularly for COVID-19 pneumonia.

Main Methods:

• Implemented a Convolutional Neural Network (CNN) using the ResNet50 architecture.
• Employed transfer learning, training the model on the COVID-CT and COVIDNet datasets.
• Integrated the LIME (Local Interpretable Model-agnostic Explanations) technique for result interpretation.

Main Results:

• Achieved high classification accuracies of 93% and 97% for pulmonary disease detection.
• Demonstrated that the XAI model highlights clinically relevant regions in radiographs, validated by radiologists.
• Successfully provided explanations for disease classification, aiding in understanding diagnostic reasoning.

Conclusions:

• The proposed XAI approach improves the accuracy of pulmonary disease classification from chest radiographs.
• The method offers interpretable explanations, assisting radiologists in clinical decision-making.
• This research advances the application of deep learning in early-stage diagnosis and treatment of lung diseases.